What are the best qualities in dumbing down?

[S.Taylor]

What should be the features which make an engine a true and measured and useful dumbing down?

1).Weak positional moves but perfect tactics?

2).Tactics which lose up to a pawn at a time?

3). Bigger tactical errors which you don't know if they are bluffs or actual give aways?

4).Allowing gifts only if you spot clever tactics?

5).Weak positional moves with....what type of tactics? Perhaps the biggest difficulty. If the tactics are too strong, so what can you do with the positional weaknesses when you are going to miss some strong tactics every other move? If they are too weak, so what have you achieved? You didn't need to use much positional understanding to achieve anything.


6). Weak positional moves with medium tactics? What's meant by medium tactics, and how would it be programmed?

[Ovyron]

I like "material missevaluation". The engine plays weak moves because it got wrong material evaluations, so it ends calculating principal variations that have nothing to do with the actual game, and sacrificing material for no reason.

For example, with Pro Deo, you can get it to exchange its queen for 7 pawns worth of material. What I like about this, is that the playing style is consistent, and the engine has a goal (A wrong goal, but a goal after all), not just one engine playing like Grand Master the full game and doing one big blunder now and then (And, the problem with other approaches is to make the weaknesses look human like.)

[hgm]

What about the following mode for dumbing down: randomly remove some moves from the move list in deeper noded, with a probability dependent on when these moves first became possible. So a move that is also in the move list for the root will almost never be removed, but a move that is absent in all move lists on the path towards a nod with depth <10 would have a pretty large probability to be removed during move generation.

I don't know if anyone ever tried this, but it seems to me that this is the way Humans make tactical mistakes. They overlook that a certain move was possible.